Article Details

Abstract

To address BMI stratification, label noise, and repeated measurements in NIPT timing of male-fetus attainment and female-fetus anomaly screening, this study analyzed single-center data using K-means clustering, a multimodal MLP, LSTM, CleanLearning, and cascade screening. For male-fetus attainment classification, the multimodal MLP achieved an AUC of 0.673, outperforming XGBoost (0.614); under patient-wise evaluation, row-wise splitting inflated the AUC from 0.673 to 0.811. In the label-noise comparison experiment without prior removal of contradictory negatives, CleanLearning improved female-fetus row-level AUC from 0.549 to 0.722; the main screening pipeline (model layer) yielded an AUC of approximately 0.691. Under BMI stratification, the 90% attainment gestational age was approximately 16.0 to 17.4 weeks (20.7 ≤ BMI < 36.3) and approximately 21.7 weeks (BMI ≥ 36.3). These results suggest that, with the available features, attainment classification and gestational-age recommendations are more clinically informative than concentration regression; patient-wise evaluation and label cleaning should be standard practice in NIPT machine learning modeling; and female-fetus performance remains limited mainly by label quality and the number of positive cases.

Keywords

non-invasive prenatal testing; multimodal fusion; label noise; cascade screening; survival analysis; Machine Learning

References

1. [1] Matin D，Mahdieh M. Silent Signals: The Expanding Role of Cell-Free DNA in Women's Health, Pregnancy, and Reproductive Medicine [J].DNA and cell biology，2026，45（5）：10445498261426174.
2. [2] Liu J，Wu M，He S J，et al. Large-scale Implementation of Non-invasive Prenatal Testing for Fetal Aneuploidies [J].Nature Health，2026：657–665.
3. [3] Valovicova K，Polak M，Ottema S，et al. Impact of Maternal Body Mass Index (BMI) on the Performance of Non-Invasive Prenatal Testing (NIPT) [J/OL].Prenatal Diagnosis，2026：70161.https://doi.org/10.1002/pd.70161.
4. [4] Hu L，Zhang Y Q，Wu Y Y，et al. K-means Clustering Based on Self-Sampling and Multi-Risk Minimization in NIPT Screening [J/OL].Journal of Artificial Intelligence Practice，2026，9（1）.https://clausiuspress.com/article/17397.html.
5. [5] Luo G H. Prediction of the Timing Selection of NIPT and Abnormality Determination of Fetus Based on Logistic Regression and Comprehensive Loss Function [J].Journal of Electronic Research and Application，2026，10（2）：159-168.
6. [6] Ding Y B，Ni K X，Fan X N，et al. A Bayesian Decision-Theoretic Optimization Model for Personalized Timing of Non-Invasive Prenatal Testing Based on Maternal BMI [J/OL].Mathematics，2026，14（3）：437.https://doi.org/10.3390/math14030437.
7. [7] Thomas L. Challenges and Ethical Issues Related to Non-Invasive Prenatal Testing (NIPT) [J].OBM Genetics，2026，10（1）：1-17.
8. [8] Uçar İ，Sariyer G，Yaprak U E，et al. Fetal Fraction and Uterine Leiomyoma Volume: New Insights From Interpretable Modeling [J/OL].Prenatal Diagnosis，2026：70162.https://obgyn.onlinelibrary.wiley.com/doi/10.1002/pd.70162.
9. [9] Irina I I，Andreea C，Zina B C，et al. Prenatal Diagnosis of Sex Chromosome Aneuploidies: A Retrospective Study Using QF-PCR, SNP-Based Chromosomal Microarray Analysis, and NIPT [J/OL].Genes，2026，17（2）：171.https://doi.org/10.3390/genes17020171.
10. [10] Pan X L，Weng L X，Pan Y，et al. Borderline Z-scores in Non-invasive Prenatal Screening: Does its Presence Hold clinical Significance? [J].Archives of Gynecology and Obstetrics，2025，312：2079-2087.
11. [11] Márton K，Ágnes C，Máté É，et al. Discordant findings in genome-wide non-invasive prenatal testing (GW-NIPT) for rare chromosomal abnormalities, adverse pregnancy outcomes, and maternal malignancies: a systematic review and meta-analysis [J].American journal of obstetrics and gynecology，2025，234（5）：1342-1354.
12. [12] Jiang Z H，Wang W J，Zhang Y，et al. High Risk for Rare Autosomal Trisomies Detected by Non‐Invasive Prenatal Testing: Prenatal Diagnosis and Pregnancy Outcomes [J/OL].Prenatal Diagnosis，2026：70181.https://obgyn.onlinelibrary.wiley.com/doi/10.1002/pd.70181.
13. [13] Wang H Y，Zhang L，Yang T，et al. Clinical Implications and Limitations of Noninvasive Prenatal Testing for Detecting Fetal Copy Number Variations: A Multicenter Study in Shaanxi Province, China [J/OL].Medical Science Monitor: international medical journal of experimental and clinical research，2026，32：e952341.https://medscimonit.com/abstract/full/idArt/952341/s/A.

About the authors

Ren Junlin (2004-), female, Miao ethnicity, from Pengshui, Chongqing, is currently pursuing a bachelor's degree. Her research interests include machine learning, medical information processing, and data mining.